Bias Mitigation
Measures to identify and reduce discriminatory outcomes in AI systems.
Definition
Bias Mitigation refers to the technical and organizational measures taken to identify, assess, and reduce biases in AI systems that could lead to discriminatory outcomes. Bias can arise from training data, algorithmic design, or the context in which AI systems are deployed.
Under the EU AI Act, providers of high-risk AI systems must implement data governance practices that examine training data for possible biases. Systems must be designed to ensure that outputs do not reflect biases that could discriminate against individuals based on protected characteristics.
Bias mitigation is an ongoing process that requires continuous monitoring, particularly in post-market surveillance, as biases may emerge or evolve over time with changing data or contexts.
Sources
- •EU AI Act Article 10
- •NIST SP 1270
Related Terms
Training Data
Data used to develop and train AI models to perform their intended functions....
Fundamental Rights Impact Assessment
Evaluation of AI system's potential impact on human rights before deployment....
Risk Management System
Continuous process to identify, analyze, and mitigate AI system risks throughout its lifecycle....